xfs_buf.c

linux-2.6.15.6

		if (!XFS_BUF_ISDONE(pb)) {
			PB_TRACE(pb, "read", (unsigned long)flags);
			XFS_STATS_INC(pb_get_read);
			pagebuf_iostart(pb, flags);
		} else if (flags & PBF_ASYNC) {
			PB_TRACE(pb, "read_async", (unsigned long)flags);
			/*
			 * Read ahead call which is already satisfied,
			 * drop the buffer
			 */
			goto no_buffer;
		} else {
			PB_TRACE(pb, "read_done", (unsigned long)flags);
			/* We do not want read in the flags */
			pb->pb_flags &= ~PBF_READ;
		}
	}

	return pb;

 no_buffer:
	if (flags & (PBF_LOCK | PBF_TRYLOCK))
		pagebuf_unlock(pb);
	pagebuf_rele(pb);
	return NULL;
}

/*
 * If we are not low on memory then do the readahead in a deadlock
 * safe manner.
 */
void
pagebuf_readahead(
	xfs_buftarg_t		*target,
	loff_t			ioff,
	size_t			isize,
	page_buf_flags_t	flags)
{
	struct backing_dev_info *bdi;

	bdi = target->pbr_mapping->backing_dev_info;
	if (bdi_read_congested(bdi))
		return;

	flags |= (PBF_TRYLOCK|PBF_ASYNC|PBF_READ_AHEAD);
	xfs_buf_read_flags(target, ioff, isize, flags);
}

xfs_buf_t *
pagebuf_get_empty(
	size_t			len,
	xfs_buftarg_t		*target)
{
	xfs_buf_t		*pb;

	pb = pagebuf_allocate(0);
	if (pb)
		_pagebuf_initialize(pb, target, 0, len, 0);
	return pb;
}

static inline struct page *
mem_to_page(
	void			*addr)
{
	if (((unsigned long)addr < VMALLOC_START) ||
	    ((unsigned long)addr >= VMALLOC_END)) {
		return virt_to_page(addr);
	} else {
		return vmalloc_to_page(addr);
	}
}

int
pagebuf_associate_memory(
	xfs_buf_t		*pb,
	void			*mem,
	size_t			len)
{
	int			rval;
	int			i = 0;
	size_t			ptr;
	size_t			end, end_cur;
	off_t			offset;
	int			page_count;

	page_count = PAGE_CACHE_ALIGN(len) >> PAGE_CACHE_SHIFT;
	offset = (off_t) mem - ((off_t)mem & PAGE_CACHE_MASK);
	if (offset && (len > PAGE_CACHE_SIZE))
		page_count++;

	/* Free any previous set of page pointers */
	if (pb->pb_pages)
		_pagebuf_free_pages(pb);

	pb->pb_pages = NULL;
	pb->pb_addr = mem;

	rval = _pagebuf_get_pages(pb, page_count, 0);
	if (rval)
		return rval;

	pb->pb_offset = offset;
	ptr = (size_t) mem & PAGE_CACHE_MASK;
	end = PAGE_CACHE_ALIGN((size_t) mem + len);
	end_cur = end;
	/* set up first page */
	pb->pb_pages[0] = mem_to_page(mem);

	ptr += PAGE_CACHE_SIZE;
	pb->pb_page_count = ++i;
	while (ptr < end) {
		pb->pb_pages[i] = mem_to_page((void *)ptr);
		pb->pb_page_count = ++i;
		ptr += PAGE_CACHE_SIZE;
	}
	pb->pb_locked = 0;

	pb->pb_count_desired = pb->pb_buffer_length = len;
	pb->pb_flags |= PBF_MAPPED;

	return 0;
}

xfs_buf_t *
pagebuf_get_no_daddr(
	size_t			len,
	xfs_buftarg_t		*target)
{
	size_t			malloc_len = len;
	xfs_buf_t		*bp;
	void			*data;
	int			error;

	bp = pagebuf_allocate(0);
	if (unlikely(bp == NULL))
		goto fail;
	_pagebuf_initialize(bp, target, 0, len, 0);

 try_again:
	data = kmem_alloc(malloc_len, KM_SLEEP | KM_MAYFAIL);
	if (unlikely(data == NULL))
		goto fail_free_buf;

	/* check whether alignment matches.. */
	if ((__psunsigned_t)data !=
	    ((__psunsigned_t)data & ~target->pbr_smask)) {
		/* .. else double the size and try again */
		kmem_free(data, malloc_len);
		malloc_len <<= 1;
		goto try_again;
	}

	error = pagebuf_associate_memory(bp, data, len);
	if (error)
		goto fail_free_mem;
	bp->pb_flags |= _PBF_KMEM_ALLOC;

	pagebuf_unlock(bp);

	PB_TRACE(bp, "no_daddr", data);
	return bp;
 fail_free_mem:
	kmem_free(data, malloc_len);
 fail_free_buf:
	pagebuf_free(bp);
 fail:
	return NULL;
}

/*
 *	pagebuf_hold
 *
 *	Increment reference count on buffer, to hold the buffer concurrently
 *	with another thread which may release (free) the buffer asynchronously.
 *
 *	Must hold the buffer already to call this function.
 */
void
pagebuf_hold(
	xfs_buf_t		*pb)
{
	atomic_inc(&pb->pb_hold);
	PB_TRACE(pb, "hold", 0);
}

/*
 *	pagebuf_rele
 *
 *	pagebuf_rele releases a hold on the specified buffer.  If the
 *	the hold count is 1, pagebuf_rele calls pagebuf_free.
 */
void
pagebuf_rele(
	xfs_buf_t		*pb)
{
	xfs_bufhash_t		*hash = pb->pb_hash;

	PB_TRACE(pb, "rele", pb->pb_relse);

	if (unlikely(!hash)) {
		ASSERT(!pb->pb_relse);
		if (atomic_dec_and_test(&pb->pb_hold))
			xfs_buf_free(pb);
		return;
	}

	if (atomic_dec_and_lock(&pb->pb_hold, &hash->bh_lock)) {
		if (pb->pb_relse) {
			atomic_inc(&pb->pb_hold);
			spin_unlock(&hash->bh_lock);
			(*(pb->pb_relse)) (pb);
		} else if (pb->pb_flags & PBF_FS_MANAGED) {
			spin_unlock(&hash->bh_lock);
		} else {
			ASSERT(!(pb->pb_flags & (PBF_DELWRI|_PBF_DELWRI_Q)));
			list_del_init(&pb->pb_hash_list);
			spin_unlock(&hash->bh_lock);
			pagebuf_free(pb);
		}
	} else {
		/*
		 * Catch reference count leaks
		 */
		ASSERT(atomic_read(&pb->pb_hold) >= 0);
	}
}


/*
 *	Mutual exclusion on buffers.  Locking model:
 *
 *	Buffers associated with inodes for which buffer locking
 *	is not enabled are not protected by semaphores, and are
 *	assumed to be exclusively owned by the caller.  There is a
 *	spinlock in the buffer, used by the caller when concurrent
 *	access is possible.
 */

/*
 *	pagebuf_cond_lock
 *
 *	pagebuf_cond_lock locks a buffer object, if it is not already locked.
 *	Note that this in no way
 *	locks the underlying pages, so it is only useful for synchronizing
 *	concurrent use of page buffer objects, not for synchronizing independent
 *	access to the underlying pages.
 */
int
pagebuf_cond_lock(			/* lock buffer, if not locked	*/
					/* returns -EBUSY if locked)	*/
	xfs_buf_t		*pb)
{
	int			locked;

	locked = down_trylock(&pb->pb_sema) == 0;
	if (locked) {
		PB_SET_OWNER(pb);
	}
	PB_TRACE(pb, "cond_lock", (long)locked);
	return(locked ? 0 : -EBUSY);
}

#if defined(DEBUG) || defined(XFS_BLI_TRACE)
/*
 *	pagebuf_lock_value
 *
 *	Return lock value for a pagebuf
 */
int
pagebuf_lock_value(
	xfs_buf_t		*pb)
{
	return(atomic_read(&pb->pb_sema.count));
}
#endif

/*
 *	pagebuf_lock
 *
 *	pagebuf_lock locks a buffer object.  Note that this in no way
 *	locks the underlying pages, so it is only useful for synchronizing
 *	concurrent use of page buffer objects, not for synchronizing independent
 *	access to the underlying pages.
 */
int
pagebuf_lock(
	xfs_buf_t		*pb)
{
	PB_TRACE(pb, "lock", 0);
	if (atomic_read(&pb->pb_io_remaining))
		blk_run_address_space(pb->pb_target->pbr_mapping);
	down(&pb->pb_sema);
	PB_SET_OWNER(pb);
	PB_TRACE(pb, "locked", 0);
	return 0;
}

/*
 *	pagebuf_unlock
 *
 *	pagebuf_unlock releases the lock on the buffer object created by
 *	pagebuf_lock or pagebuf_cond_lock (not any pinning of underlying pages
 *	created by pagebuf_pin).
 *
 *	If the buffer is marked delwri but is not queued, do so before we
 *	unlock the buffer as we need to set flags correctly. We also need to
 *	take a reference for the delwri queue because the unlocker is going to
 *	drop their's and they don't know we just queued it.
 */
void
pagebuf_unlock(				/* unlock buffer		*/
	xfs_buf_t		*pb)	/* buffer to unlock		*/
{
	if ((pb->pb_flags & (PBF_DELWRI|_PBF_DELWRI_Q)) == PBF_DELWRI) {
		atomic_inc(&pb->pb_hold);
		pb->pb_flags |= PBF_ASYNC;
		pagebuf_delwri_queue(pb, 0);
	}

	PB_CLEAR_OWNER(pb);
	up(&pb->pb_sema);
	PB_TRACE(pb, "unlock", 0);
}


/*
 *	Pinning Buffer Storage in Memory
 */

/*
 *	pagebuf_pin
 *
 *	pagebuf_pin locks all of the memory represented by a buffer in
 *	memory.  Multiple calls to pagebuf_pin and pagebuf_unpin, for
 *	the same or different buffers affecting a given page, will
 *	properly count the number of outstanding "pin" requests.  The
 *	buffer may be released after the pagebuf_pin and a different
 *	buffer used when calling pagebuf_unpin, if desired.
 *	pagebuf_pin should be used by the file system when it wants be
 *	assured that no attempt will be made to force the affected
 *	memory to disk.	 It does not assure that a given logical page
 *	will not be moved to a different physical page.
 */
void
pagebuf_pin(
	xfs_buf_t		*pb)
{
	atomic_inc(&pb->pb_pin_count);
	PB_TRACE(pb, "pin", (long)pb->pb_pin_count.counter);
}

/*
 *	pagebuf_unpin
 *
 *	pagebuf_unpin reverses the locking of memory performed by
 *	pagebuf_pin.  Note that both functions affected the logical
 *	pages associated with the buffer, not the buffer itself.
 */
void
pagebuf_unpin(
	xfs_buf_t		*pb)
{
	if (atomic_dec_and_test(&pb->pb_pin_count)) {
		wake_up_all(&pb->pb_waiters);
	}
	PB_TRACE(pb, "unpin", (long)pb->pb_pin_count.counter);
}

int
pagebuf_ispin(
	xfs_buf_t		*pb)
{
	return atomic_read(&pb->pb_pin_count);
}

/*
 *	pagebuf_wait_unpin
 *
 *	pagebuf_wait_unpin waits until all of the memory associated
 *	with the buffer is not longer locked in memory.  It returns
 *	immediately if none of the affected pages are locked.
 */
static inline void
_pagebuf_wait_unpin(
	xfs_buf_t		*pb)
{
	DECLARE_WAITQUEUE	(wait, current);

	if (atomic_read(&pb->pb_pin_count) == 0)
		return;

	add_wait_queue(&pb->pb_waiters, &wait);
	for (;;) {
		set_current_state(TASK_UNINTERRUPTIBLE);
		if (atomic_read(&pb->pb_pin_count) == 0)
			break;
		if (atomic_read(&pb->pb_io_remaining))
			blk_run_address_space(pb->pb_target->pbr_mapping);
		schedule();
	}
	remove_wait_queue(&pb->pb_waiters, &wait);
	set_current_state(TASK_RUNNING);
}

/*
 *	Buffer Utility Routines
 */

/*
 *	pagebuf_iodone
 *
 *	pagebuf_iodone marks a buffer for which I/O is in progress
 *	done with respect to that I/O.	The pb_iodone routine, if
 *	present, will be called as a side-effect.
 */
STATIC void
pagebuf_iodone_work(
	void			*v)
{
	xfs_buf_t		*bp = (xfs_buf_t *)v;

	if (bp->pb_iodone)
		(*(bp->pb_iodone))(bp);
	else if (bp->pb_flags & PBF_ASYNC)
		xfs_buf_relse(bp);
}

void
pagebuf_iodone(
	xfs_buf_t		*pb,
	int			schedule)
{
	pb->pb_flags &= ~(PBF_READ | PBF_WRITE);
	if (pb->pb_error == 0)
		pb->pb_flags |= PBF_DONE;

	PB_TRACE(pb, "iodone", pb->pb_iodone);

	if ((pb->pb_iodone) || (pb->pb_flags & PBF_ASYNC)) {
		if (schedule) {
			INIT_WORK(&pb->pb_iodone_work, pagebuf_iodone_work, pb);
			queue_work(xfslogd_workqueue, &pb->pb_iodone_work);
		} else {
			pagebuf_iodone_work(pb);
		}
	} else {
		up(&pb->pb_iodonesema);
	}
}

/*
 *	pagebuf_ioerror
 *
 *	pagebuf_ioerror sets the error code for a buffer.
 */
void
pagebuf_ioerror(			/* mark/clear buffer error flag */
	xfs_buf_t		*pb,	/* buffer to mark		*/
	int			error)	/* error to store (0 if none)	*/
{
	ASSERT(error >= 0 && error <= 0xffff);
	pb->pb_error = (unsigned short)error;
	PB_TRACE(pb, "ioerror", (unsigned long)error);
}

/*
 *	pagebuf_iostart
 *
 *	pagebuf_iostart initiates I/O on a buffer, based on the flags supplied.
 *	If necessary, it will arrange for any disk space allocation required,
 *	and it will break up the request if the block mappings require it.
 *	The pb_iodone routine in the buffer supplied will only be called
 *	when all of the subsidiary I/O requests, if any, have been completed.
 *	pagebuf_iostart calls the pagebuf_ioinitiate routine or
 *	pagebuf_iorequest, if the former routine is not defined, to start
 *	the I/O on a given low-level request.
 */
int
pagebuf_iostart(			/* start I/O on a buffer	  */
	xfs_buf_t		*pb,	/* buffer to start		  */
	page_buf_flags_t	flags)	/* PBF_LOCK, PBF_ASYNC, PBF_READ, */
					/* PBF_WRITE, PBF_DELWRI,	  */
					/* PBF_DONT_BLOCK		  */
{
	int			status = 0;

	PB_TRACE(pb, "iostart", (unsigned long)flags);

	if (flags & PBF_DELWRI) {
		pb->pb_flags &= ~(PBF_READ | PBF_WRITE | PBF_ASYNC);
		pb->pb_flags |= flags & (PBF_DELWRI | PBF_ASYNC);
		pagebuf_delwri_queue(pb, 1);
		return status;
	}

	pb->pb_flags &= ~(PBF_READ | PBF_WRITE | PBF_ASYNC | PBF_DELWRI | \
			PBF_READ_AHEAD | _PBF_RUN_QUEUES);
	pb->pb_flags |= flags & (PBF_READ | PBF_WRITE | PBF_ASYNC | \
			PBF_READ_AHEAD | _PBF_RUN_QUEUES);

	BUG_ON(pb->pb_bn == XFS_BUF_DADDR_NULL);

	/* For writes allow an alternate strategy routine to precede
	 * the actual I/O request (which may not be issued at all in
	 * a shutdown situation, for example).
	 */
	status = (flags & PBF_WRITE) ?
		pagebuf_iostrategy(pb) : pagebuf_iorequest(pb);

	/* Wait for I/O if we are not an async request.
	 * Note: async I/O request completion will release the buffer,
	 * and that can already be done by this point.  So using the
	 * buffer pointer from here on, after async I/O, is invalid.
	 */
	if (!status && !(flags & PBF_ASYNC))
		status = pagebuf_iowait(pb);

	return status;
}

/*
 * Helper routine for pagebuf_iorequest
 */

STATIC __inline__ int
_pagebuf_iolocked(
	xfs_buf_t		*pb)
{
	ASSERT(pb->pb_flags & (PBF_READ|PBF_WRITE));
	if (pb->pb_flags & PBF_READ)
		return pb->pb_locked;
	return 0;
}

STATIC __inline__ void
_pagebuf_iodone(
	xfs_buf_t		*pb,
	int			schedule)
{
	if (atomic_dec_and_test(&pb->pb_io_remaining) == 1) {
		pb->pb_locked = 0;
		pagebuf_iodone(pb, schedule);
	}
}

STATIC int
bio_end_io_pagebuf(
	struct bio		*bio,
	unsigned int		bytes_done,
	int			error)
{
	xfs_buf_t		*pb = (xfs_buf_t *)bio->bi_private;
	unsigned int		blocksize = pb->pb_target->pbr_bsize;
	struct bio_vec		*bvec = bio->bi_io_vec + bio->bi_vcnt - 1;

	if (bio->bi_size)
		return 1;

	if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
		pb->pb_error = EIO;

	do {
		struct page	*page = bvec->bv_page;

		if (unlikely(pb->pb_error)) {
			if (pb->pb_flags & PBF_READ)
				ClearPageUptodate(page);
			SetPageError(page);
		} else if (blocksize == PAGE_CACHE_SIZE) {
			SetPageUptodate(page);
		} else if (!PagePrivate(page) &&
				(pb->pb_flags & _PBF_PAGE_CACHE)) {
			set_page_region(page, bvec->bv_offset, bvec->bv_len);
		}

		if (--bvec >= bio->bi_io_vec)
			prefetchw(&bvec->bv_page->flags);

		if (_pagebuf_iolocked(pb)) {
			unlock_page(page);
		}
	} while (bvec >= bio->bi_io_vec);

	_pagebuf_iodone(pb, 1);
	bio_put(bio);
	return 0;
}

STATIC void
_pagebuf_ioapply(
	xfs_buf_t		*pb)
{
	int			i, rw, map_i, total_nr_pages, nr_pages;
	struct bio		*bio;
	int			offset = pb->pb_offset;
	int			size = pb->pb_count_desired;
	sector_t		sector = pb->pb_bn;
	unsigned int		blocksize = pb->pb_target->pbr_bsize;
	int			locking = _pagebuf_iolocked(pb);

	total_nr_pages = pb->pb_page_count;
	map_i = 0;

	if (pb->pb_flags & _PBF_RUN_QUEUES) {
		pb->pb_flags &= ~_PBF_RUN_QUEUES;
		rw = (pb->pb_flags & PBF_READ) ? READ_SYNC : WRITE_SYNC;
	} else {
		rw = (pb->pb_flags & PBF_READ) ? READ : WRITE;
	}

	if (pb->pb_flags & PBF_ORDERED) {
		ASSERT(!(pb->pb_flags & PBF_READ));
		rw = WRITE_BARRIER;
	}

	/* Special code path for reading a sub page size pagebuf in --
	 * we populate up the whole page, and hence the other metadata
	 * in the same page.  This optimization is only valid when the
	 * filesystem block size and the page size are equal.
	 */
	if ((pb->pb_buffer_length < PAGE_CACHE_SIZE) &&
	    (pb->pb_flags & PBF_READ) && locking &&
	    (blocksize == PAGE_CACHE_SIZE)) {
		bio = bio_alloc(GFP_NOIO, 1);

		bio->bi_bdev = pb->pb_target->pbr_bdev;
		bio->bi_sector = sector - (offset >> BBSHIFT);
		bio->bi_end_io = bio_end_io_pagebuf;
		bio->bi_private = pb;

		bio_add_page(bio, pb->pb_pages[0], PAGE_CACHE_SIZE, 0);